Profiling the propagation of error from PPG to HRV features in a wearable

physiological-monitoring device
Davide Morelli1,2,3 ✉, Leonardo Bartoloni1,2, Michele Colombo1,2, David Plans1,3, David A. Clifton4
1
BioBeats Group Ltd, London, UK
2
Dipartimento di Informatica, Università di Pisa, Pisa, Italy
3
Center for Digital Economy, University of Surrey, Guildford, UK
4
Department of Engineering Science, University of Oxford, Oxford, UK
✉ E-mail: [email protected]

Published in Healthcare Technology Letters; Received on 24th May 2017; Revised on 18th July 2017; Accepted on 19th July 2017

Wearable physiological monitors are becoming increasingly commonplace in the consumer domain, but in literature there exists no substantive
studies of their performance when measuring the physiology of ambulatory patients. In this Letter, the authors investigate the reliability of
the heart-rate (HR) sensor in an exemplar ‘wearable’ wrist-worn monitoring system (the Microsoft Band 2); their experiments quantify the
propagation of error from (i) the photoplethysmogram (PPG) acquired by pulse oximetry, to (ii) estimation of HR, and (iii) subsequent
calculation of HR variability (HRV) features. Their experiments confirm that motion artefacts account for the majority of this error, and
show that the unreliable portions of HR data can be removed, using the accelerometer sensor from the wearable device. The experiments
further show that acquired signals contain noise with substantial energy in the high-frequency band, and that this contributes to
subsequent variability in standard HRV features often used in clinical practice. The authors finally show that the conventional use of long-
duration windows of data is not needed to perform accurate estimation of time-domain HRV features.

1. Introduction: Wearable physiological monitoring, exploiting features as movement of the patient increases; and (iii) suggest a
devices with unobtrusive (often wrist-worn) packages, offers means of filtering to minimise the error of the HRV features.
substantial promise for improving the care of patients in clinical
settings, and for enabling individuals to better manage their own
1.1. Heart-rate variability: HRV is the beat-to-beat variability in
health. With recent advances in consumer markets, including
R–R intervals (sometimes termed NN intervals or successive
devices such as fitness trackers and ‘smart watches’, the use
differences (SD) in the literature), where the latter refers to the time
of wearable monitors is becoming increasingly commonplace.
series obtained by identifying the duration between subsequent
However, very few of these devices penetrate into use at scale
R-peaks in the electrocardiogram (ECG) waveform. An R–R
within either clinical settings or for permitting patients to track
interval t therefore corresponds to an estimate of instantaneous HR
their own health outside clinical environments [1, 2], with only
60t −1 bpm (beats per minute). HRV features are descriptors that
small numbers of studies that have been described in the literature
capture aspect of this variability in the time or frequency domain.
[3–5]. A major obstacle to the use of wearable devices in such
Examples of the former include the following, which are defined
settings is the lack of characterisation of their ability to estimate
for a window of data of duration t, where often t = 5 min:
clinically relevant physiological parameters, such as, in the case
of cardiac applications, understanding the propagation of error
from the photoplethysmogram (PPG) waveform acquired within † AverageNN, defined as being the mean of the NN intervals
the device, through to estimation of the heart rate (HR), and occurring within the window [4, 11];
ultimately on to the heart-rate variability (HRV) features that are † SDNN, defined as being the standard deviation of the NN inter-
used to track the state of cardiac health of a patient. vals occurring within the window [4, 11–13];
In this Letter, we investigate this propagation of error, with the † RMSSD, the root-mean-square (RMS) of SD occurring within
aim of improving understanding of the accuracy of the estimated the window [4, 11, 13];
HRV features; we use as our experimental device a commonly † pNN50, which is the proportion of the total number of NN inter-
used wearable consumer device, the Microsoft Band 2. The most vals within the window that exceed 50 ms. That is, one first defines
commonly occurring mode of error in such wearable devices has {NN50} as being that subset of {NN intervals} that exceed a dur-
previously been identified as being due to movement artefact [6], ation of 50 ms, and where pNN50 is |NN50|/|NNintervals|, where
which is unsurprising given the sensitivity of the pulse oximetry | · | is set cardinality [4, 11, 13].
process (which yields the PPG waveform from which HR and
HRV parameters are estimated) to the small changes in light inten- Examples of frequency-domain HRV features include:
sity that arise due to movement of the patient. However, while some
studies have investigated how to mitigate the effects of motion arte- † SVI, which is the sympathovagal index, defined as the LF:HF
facts correcting the raw PPG signal [7–9], we could find a small ratio, where LF is the total power in the low-frequency band
amount of studies investigating the possibility to assess PPG reli- (between 0.003 and 0.14 Hz), and where HF is the total power
ability from accelerometer [10], and no study investigating the in the high-frequency band (between 0.15 and 0.40 Hz). The defin-
propagation of error from PPG to HRV features. ition of the boundaries of the LF and HF bands varies slightly
This investigation provides the following contributions: (i) inves- between authors, being a heuristic definition [11, 12, 14].
tigation of a means of discriminating when the time series of pulsa-
tile intervals (estimated from the PPG) should be discarded; For simplicity, we use the same name for HRV features cal-
(ii) estimate the expected error on subsequently derived HRV culated from R–R intervals (collected from the ECG sensor), and

Healthcare Technology Letters, 2018, Vol. 5, Iss. 2, pp. 59–64 59

doi: 10.1049/htl.2017.0039 This is an open access article published by the IET under the
Creative Commons Attribution -NonCommercial License (http://
creativecommons.org/licenses/by-nc/3.0/)
the HRV features estimated from P–P intervals (collected from the 4. For t = [1200 1800] s, the subject is requested to remain sitting,
PPG sensor). but is permitted to perform normal seated activities; e.g. talking,
HRV analysis is traditionally carried out via ECG collected in a using a laptop computer, and so on.
controlled environment; analysis thereby proceeds on the assump-
tion of noise-free data. Many HRV features aim to capture fast vari-
ability (high frequencies) in R–R time series, which is easily and
adversely affected by the presence of outliers. For this reason, the 3. Methods: This section describes the steps taken in analysing the
time series of RR intervals is typically reviewed manually, before data acquired from each subject. Aggregated results for all subjects
their use in analysis. will be presented later, in Section 4.
The use of PPG instead of ECG for HRV analysis is not exten-
sively explored, and especially for those cases in which the PPG 3.1. Accelerometry: From the three time series of accelerometry xt ,
is acquired via consumer-grade wearable devices. Various studies yt , zt , corresponding to motions in each of three orthogonal
have explored the accuracy of HRV features extracted from clinic- directions, we consider absolute magnitude of the first-order
ally graded PPG devices [13, 15–20], while analysis of the quality differences, which is therefore invariant to rotation of the device
of HRV from a PPG acquired from smartphones has been described
[21]. These studies have, in general, concluded that HRV analysis 











































from PPG acquired under carefully controlled conditions is reliable, wt = (xt − xt−1 )2 + (yt − yt−1 )2 + (zt − zt−1 )2 (1)
because most HRV features estimated via PPG show a very high
correlation with the same features estimated via ECG. A notable ex- which are subsequently averaged over a window of duration t,
ception to the latter exists for pNN50 [13, 17], which was shown to containing n samples
be less reliable when estimated via the PPG than from the ECG.
It has been reported that use of the PPG is only reliable if the 1 t

cardiovascular system is in a stationary condition [20]; i.e. that the Wt (n) = w (2)
n i=t−n i
PPG cannot reliably be used using conventional methods when the
subject is recovering from physical activity. In our experiment, The time series wt is shown for an exemplar subject in Fig. 1. It may
we investigate this finding (and ultimately find no difference in accur- be seen from the figure that the accelerometry time series increases
acy after a short period of exercise than when compared with PPG with the activity undertaken during the second phase of the
from a resting condition). We here describe our investigation of the experiment (in which the subject was climbing and descending
accuracy of HRV features estimated from a wrist-worn PPG sensor stairs), and that there was very little activity during the first
in a non-controlled environment, as would be represented by a and third sections (during which the subject remained still). The
typical setting for the use of wearable devices. This is a crucial pre- ‘free-action’ fourth section shows little activity, with occasional
liminary task in HRV analysis from wearable devices, and, to our transients in activity, arising from rapid and infrequent
knowledge, no study has yet been carried out on this subject. movements of the wrist on which the device was worn.
To our knowledge this is the first analysis of the error propaga- The figure also shows that there exist the expected differences
tion from P–P intervals to HRV features. between P–P and R–R intervals estimated from the PPG and ECG
sensors. During the first and third sections, the time series of P–P
and R–R intervals are similar. As motion increases, it may be seen
2. Experimental design: The trial from which data was derived for that the fourth section corresponds to frequent substantive differences
use in this manuscript was reviewed by and received a favourable between P–P and R–R intervals, and that these differences become
ethical opinion from University of Surrey Ethics Committee, with larger during the increased activity of the second section.
reference number UEC/2016/027/FASS.
We collected 30 min of data from the wearable device for each of
3.2. Outlier removal: Before subsequent processing we performed
five subjects, all male, healthy, Fitzpatrick scale Type III, mean age
outlier removal to discard P–P intervals with physiologically
32 (standard deviation 6). These data include (i) a time series of the
implausible values, e.g. increased P–P intervals (and therefore
accelerometry for each of three orthogonal axes of motion and (ii) a
HR) that occur faster than could be feasibly produced by
time series of pulse-to-pulse intervals, both of which are estimated
physiology. This was straightforwardly performed by computing
by the band from its PPG sensor. We concurrently collected R–R
intervals using an ECG sensor (acting as the gold standard) from
a Polar H7 chest-mounted strap. We will hereafter use the term
P–P intervals to refer to both the pulse-to-pulse intervals (acquired
from the wrist-worn wearable PPG device), and R–R intervals to
refer to beat-to-beat intervals (acquired from the chest-mounted
ECG system).
Our experimental protocol comprises four sections:

1. For t = [0 300] s, the subject was requested to sit as still as pos-

sible, refraining from talking or moving the arm on which the wear-
able PPG system was mounted. This first section of the experiment
is aimed at establishing a reliable baseline for both motion artefacts
and resting HR.
2. For t = [300 600] s, the subject was requested to climb and then
descend two storeys of stairs without stopping. This second section
is aimed at measuring the error induced by motion artefacts, and to
raise substantially the HR of the subject.
3. For t = [600 1200] s, the subject was again requested to sit as still
Fig. 1 Time series of P–P and R–R intervals acquired from PPG and ECG
as possible. The aim of this third section is to establish the propa- (upper plot), and corresponding accelerometry time series wt (lower plot)
gation of error from a PPG sensor, in absence of movement, for an exemplar patient, where divisions between the four sections of the
when the subject is not in a stationary cardiac situation. experimental protocol are shown as vertical dashed lines

60 Healthcare Technology Letters, 2018, Vol. 5, Iss. 2, pp. 59–64

This is an open access article published by the IET under the doi: 10.1049/htl.2017.0039
Creative Commons Attribution -NonCommercial License (http://
creativecommons.org/licenses/by-nc/3.0/)
Fig. 2 P–P intervals selected for removal (vertical red lines), and corres-
ponding distance of instantaneous P–P interval from the 10 s moving
average, as a percentage. The blue horizontal dotted line indicates the
threshold for the distance to trigger an outlier detection

a 10 s moving average m10 , and then discarding P–P intervals PPt

for which |PPt − m10 | ≥ 0.5m10 . Fig. 2 shows this pre-processing
step applied to the R–R intervals obtained from the PPG, where it
may be seen that the largest transients are removed.

3.3. HRV analysis: We estimate the various HRV features defined

earlier in Section 1.1, which are AverageNN, SDNN, RMSSD,
pNN50, and SVI. These are estimated, together with Wt , over
a short-time window of duration t = 40 s, with subsequent
windows overlapped by 20 s.
Fig. 3 shows HRV features estimated for an exemplar subject.
During the second section of the experiment, it may be seen that
the increased movement of the subject results in substantial errors
in the PPG signal that propagate to the HRV features.
For this exemplar patient, we can see that the standard deviation
of the accelerometry time series Wt is correlated to the magnitude of
the errors in the time-domain HRV features, with correlation coef-
ficients: r = 0.91, 0.84, 0.67, and 0.64 for AverageNN, SDNN,
pNN50, and RMSSD, respectively. However, this correlation
does not exist for the frequency-domain HRV feature, SVI, for
which the corresponding correlation coefficient is r = −0.03. We
will investigate this latter phenomenon in a subsequent section of
this Letter. Fig. 3 HRV features estimated from P–P intervals acquired from PPG (red)
and from R–R intervals acquired from ECG (black), in reading order. The
lower plot shows the time series of accelerometry Wt , with mean and one SD
3.4. Discarding data with low signal quality: Having confirmed values within a window of t = 40 s shown in black and red, respectively
above the intuition that error in HRV feature estimation is
correlated with movement, we subsequently choose to estimate
signal quality via the accelerometry time series Wt , such that conducted on a small patient group, and which would naturally
HRV estimates may be discarded if Wt exceeds some threshold k. be more principled in formulation for a subsequent larger study.
We therefore calculated the distribution of error for each HRV
feature, as a function of the value of Wt (the average of wt over
the window of duration t), and noticed that error for most of the 3.5. Signal realignment: The final step in our pre-processing is to
HRV features (AverageNN, SDNN, RMSSD, and pNN50) find (and correct for) any time offset that might be between the
quickly increases for Wt . 0.02: the error for AverageNN goes PPG and ECG. This is straightforwardly performed by comparing
from 0.0013 (Wt , 0.02) to 0.0142 (Wt . 0.02); the error for the pre-processed time series of P–P intervals and R–R intervals,
SDNN goes from 0.0061 (Wt , 0.02) to 0.0139 (Wt . 0.02); the and calculating the squared distance d between the two for a
error for RMSSD goes from 0.0079 (Wt , 0.02) to 0.0153 varying time offset a. We note that the P–P and R–R intervals
(Wt . 0.02); the error for RMSSD goes from 0.057 (Wt , 0.02) are not sampled at corresponding times, and therefore both
to 0.125 (Wt . 0.02). SVI is the only HRV feature whose error waveforms were resampled at 10 Hz.
does not change with Wt . 0.02. Following the observation that, Fig. 4 shows this relationship for an exemplar patient, which is
for most of the HRV features, the error is an order of magnitude reproduced across other subjects (not shown here for brevity). We
lower than the value of the corresponding HRV feature for choose a value of a = 2.2 s, which corresponds to the value that
Wt , 0.02, and that the error becomes of the same order of magni- minimises d.
tude as the value of the HRV feature for Wt . 0.02, we choose a
threshold of k = 0.02 for the remainder of this work. We emphasise 4. Results: This section presents results of (i) investigating the
that this is a candidate k, given the prototype nature of the study, effect of changing window size t when calculating HRV features;

Healthcare Technology Letters, 2018, Vol. 5, Iss. 2, pp. 59–64 61

doi: 10.1049/htl.2017.0039 This is an open access article published by the IET under the
Creative Commons Attribution -NonCommercial License (http://
creativecommons.org/licenses/by-nc/3.0/)
 Additionally, we can plot the SNR between RR (the ‘signal’) and
rt = RR − PP (the ‘noise’) for the cases with and without pre-
processing of PP, as shown in Fig. 6.
Figs. 5 and 6 show that the power in the signal exceeds that of the
noise for the ultra-low frequency band (i.e. those frequencies below
LF). This is an expected result, because power in this band corres-
ponds to slowly changing physiological phenomena that modulate
the time series of P–P and R–R intervals (corresponding to the
HR) – this corresponds to diurnal changes in blood pressure, tem-
perature regulation, and other effects.
Fig. 4 Squared distance d between pre-processed P–P and R–R intervals, Both the LF and HF bands have a very low SNR. The LF band
as a function of time offset a between them (seconds) frequently exceeds SNR = 1.0 for the case in which the P–P inter-
vals from the PPG have been pre-processed. HF band has a SNR
close to, or below, SNR = 1.0, which corresponds to the fact that
(ii) investigating the signal-to-noise ratio (SNR) for the time series the noise is at least as powerful as the signal in this band. We
of P–P interval estimated from the PPG, with respect to the note also that the absolute power in the HF band is low with
reference R–R intervals derived from the ECG; and (iii) the respect to the lower-frequency bands (as shown in Fig. 5). This
distribution of errors for HRV features throughout the four stages latter effect may explain why SVI is as unreliable as has been
of the experimental protocol. observed by our results: noting that SVI is the ratio between LF
and HF, with HF in the denominator, it is therefore sensitive to
4.1. Effect of window size: Table 1 shows the root mean-square HF noise. We may observe that the pre-processed (‘filtered’) PPG
error (RMSE) of each HRV feature for various window sizes t. It has a higher SNR than the unprocessed (‘unfiltered’) PPG in the
may be seen that the RMSE does not change substantially with HF band. We conclude that SVI is not a reliable HRV feature
varying tau; errors increase with window size for AverageNN and should be used with caution, ensuring our pre-processing
(≃ 10%), RMSSD (≃ 10%), and pNN50 (≃ 2%), while the steps are performed.
RMSE decreases for SDNN (≃ 25%) and SVI (≃ 30%). Fig. 7 shows the distribution of SNRs for HF and LF bands
However, it may be seen from the table that the decrease of SVI across all subjects, where we emphasise that these distributions
is not proportional to changes in t. are normalised histograms. The figure shows that HF values fre-
It is common practice to require long-duration windows of data to quently fall below the value of SNR = 1 (i.e. that part of the distri-
estimate HRV features reliably. Our experiment shows that short- bution shown in black that lies to the left of the vertical dashed line).
This indicates that HF values are not reliable for the majority of
duration windows of data still allow to perform accurate estimation
of time-domain HRV features. patients. LF values appear to be more reliable than HF, since a
larger proportion of the distribution shown in red falls above
SNR = 1.
4.2. Signal-to-noise ratio: Noting that the ECG-derived time series
of R–R intervals RR is used as the reference, we can define the time
series of residuals rt = RR − PP, where the time series of
re-aligned P–P intervals from the PPG is PP.
Fig. 5 shows the spectra of RR compared with the residual time
series rt when calculated with and without pre-processing of the
time series of P–P intervals derived from the PPG.

Table 1 RMSE of HRV features, for window size t

t, s AverageNN SDNN RMSSD pNN50 SVI

Fig. 6 SNR between RR (signal) and residuals rt (noise), with and without
pre-processing of PP shown in green and black, respectively. A horizontal
40 0.0480 0.0334 0.0564 0.2285 0.6471 line shows SNR = 1.0
60 0.0489 0.0316 0.0566 0.2294 0.5683
80 0.0499 0.0294 0.0575 0.2279 0.6080
100 0.0504 0.0287 0.0583 0.2330 0.5982
120 0.0506 0.0280 0.0575 0.2259 0.6613

Fig. 5 Spectra for the signal RR versus residuals rt where the latter have
been calculated from the original (‘noise’) and pre-processed (‘noise fil-
tered’) signals PP, shown in red and green, respectively. The vertical Fig. 7 Distribution of SNR in HF and LF bands for all subjects. SNR = 1.0
lines show the frequency bands used in HRV analysis: LF between 0.04 is shown with the vertical dashed line; values to the left of this line corres-
and 0.15 Hz, and HF between 0.15 and 0.4 Hz pond to a proportion of subjects’ data that falls below SNR = 1.0

62 Healthcare Technology Letters, 2018, Vol. 5, Iss. 2, pp. 59–64

This is an open access article published by the IET under the doi: 10.1049/htl.2017.0039
Creative Commons Attribution -NonCommercial License (http://
creativecommons.org/licenses/by-nc/3.0/)
4.3. Distributions of HRV errors for all subjects: In the previous Table 6 Distribution of pNN50 error, showing the mean and SD across
sections, we applied the accelerometer-based noise filtering and all subjects, and the quantiles on the distribution of pNN50 error at 0.1,
subsequent HRV analysis to the data collected from a single user. 0.25, 0.75, and 0.9
In this section, we replicate the same procedure to the data
collected from all users. For every HRV feature we aggregate the Mean SD 0.25 0.5 0.75 0.9
results and we calculate the mean, the standard deviation, the
25th, 50th, and 75th percentiles. We present those results as rest 0.238 0.107 0.167 0.235 0.293 0.377
stress 0.511 0.086 0.438 0.525 0.565 0.634
statistical descriptors of the expected error propagation from PPG
recovery 0.021 0.049 0.000 0.000 0.030 0.060
to HRV features.
free 0.174 0.116 0.076 0.165 0.260 0.332
Table 2 shows the distribution of accelerometry Wt over all sub-
jects in the four stages of the experimental protocol (rest, stress, re-
covery, and free). As expected, the rest and recovery stages have
similar accelerometry, with more than 75% of the Wt values Table 7 Distribution of SVI error, showing the mean and SD across all
subjects, and the quantiles on the distribution of SVI error at 0.1, 0.25,
below our threshold k value of Wt = 0.02). During the free stage,
0.75, and 0.9
values of Wt fall below 0.02 than 50% of the time. During the
stress stage, Wt consistently exceeds the threshold k. Mean SD 0.25 0.5 0.75 0.9
Tables 3–6 show the distribution of errors in the time-domain
HRV features, for each phase. Reflecting the distribution of the rest 1.536 1.298 0.745 1.022 2.099 2.377
accelerometry Wt , the rest and recovery stages have low error stress 0.611 0.535 0.147 0.451 0.772 1.515
most of the time (75%), the free stage has low error between recovery 0.659 0.452 0.332 0.537 0.932 1.390
free 1.308 0.994 0.512 1.112 2.057 2.948

Table 2 Distribution of accelerometry Wt , showing the mean and standard

deviation (SD) across all subjects, and the quantiles on the distribution of 25 and 50% of the time, and the stress stage is consistently asso-
Wt at 0.1, 0.25, 0.75, and 0.9 ciated with large error in HRV features.
Table 7 shows the error associated with the frequency-domain
Mean SD 0.25 0.5 0.75 0.9
HRV feature, SVI. As concluded previously, this feature appears
to be generally less reliable than the time-domain HRV features,
rest 0.016 0.032 0.002 0.002 0.010 0.072
across all subjects. The large errors present in the estimation of
stress 0.406 0.202 0.212 0.471 0.577 0.600
recovery 0.012 0.035 0.002 0.003 0.007 0.011
SVI, reported in Table 7, are due to the low SNR of the HF band.
free 0.052 0.032 0.027 0.057 0.075 0.089
5. Conclusions: We have shown that PPG data acquired from
a consumer-grade wrist-worn wearable device are highly
susceptible to motion artefacts. We have analysed the noise
Table 3 Distribution of AverageNN error, showing the mean and SD profile, and shown that the frequency-domain HRV feature SVI is
across all subjects, and the quantiles on the distribution of AverageNN
not reliable, because the SNR in the HF band (which is in the
error at 0.1, 0.25, 0.75, and 0.9
denominator of the quotient that defines SVI) often falls below a
Mean SD 0.25 0.5 0.75 0.9
value of SNR = 1.0.
Our results demonstrate that HRV features extracted when the
rest 0.004 0.010 0.001 0.001 0.002 0.008 user is still reliable, even if the user is not in a cardiovascular
stress 0.127 0.052 0.088 0.124 0.155 0.194 stationary state (such as in the third stage of our experimental proto-
recovery 0.007 0.020 0.001 0.001 0.003 0.011 col, ‘recovery’), which therefore does not agree with the findings
free 0.043 0.032 0.009 0.042 0.072 0.082 described in the literature [20].
We have shown that it is possible to use the accelerometry Wt
derived from the wearable device to estimate the quality of the
corresponding PPG signal. Therefore, unreliable data can auto-
Table 4 Distribution of SDNN error, showing the mean and SD across all
matically be discarded according to a threshold k on Wt . Such an
subjects, and the quantiles on the distribution of SDNN error at 0.1, 0.25,
operation could be used to selectively turn off the PPG sensor,
0.75, and 0.9
thereby saving battery life on the wearable device, and also avoid-
Mean SD 0.25 0.5 0.75 0.9 ing HRV analysis using unreliable data. We derive a value of k that
retain data with error an order of magnitude lower than the signal
rest 0.012 0.011 0.004 0.009 0.016 0.022 for most HRV features, discarding data with higher error. The
stress 0.132 0.036 0.111 0.133 0.166 0.179 value of k should be refined with a larger dataset.
recovery 0.006 0.014 0.001 0.003 0.006 0.015 We have shown that the conventional use of long-duration
free 0.030 0.022 0.009 0.028 0.049 0.061 windows of data is not needed to perform accurate estimation of
time-domain HRV features.
We emphasise that the results described in this report represent a
preliminary study, and planned future work includes validation of
Table 5 Distribution of RMSSD error, showing the mean and SD across
these results with more subjects to verify our findings.
all subjects, and the quantiles on the distribution of RMSSD error at 0.1,
0.25, 0.75, and 0.9
6. Funding and declaration of interests: Dr Clifton reports
Mean SD 0.25 0.5 0.75 0.9 personal fees from BioBeats Ltd. during the conduct of the study.

rest 0.025 0.014 0.016 0.024 0.031 0.038 7 References

stress 0.078 0.022 0.064 0.079 0.089 0.107
recovery 0.008 0.011 0.003 0.004 0.008 0.019 [1] Clifford G.D., Clifton D.A.: ‘Annual review: wireless technology in
free 0.033 0.020 0.017 0.034 0.049 0.056 disease state management and medicine’, Annu. Rev. Med., 2012,
63, pp. 479–492

Healthcare Technology Letters, 2018, Vol. 5, Iss. 2, pp. 59–64 63

doi: 10.1049/htl.2017.0039 This is an open access article published by the IET under the
Creative Commons Attribution -NonCommercial License (http://
creativecommons.org/licenses/by-nc/3.0/)
 [2] Tarassenko L., Clifton D.A.: ‘Semiconductor wireless technology for [12] Berntson G.G., Thomas Bigger J., Eckberg D.L., ET AL.: ‘ Heart rate
chronic disease management’, Electron. Lett., 2011, S30, pp. 30–32 variability: origins, methods, and interpretive caveats’,
[3] Clifton D.A., Pimentel M.A.F., Niehaus K., ET AL.: ‘Intelligent Psychophysiology, 1997, 34, (6), pp. 623–648
electronic health systems’, in Eren H., Webster J.G. (Eds.): [13] Jeyhani V., Mahdiani S., Peltokangas M., ET AL.: ‘Comparison of HRV
‘Telemedicine and electronic medicine’ (CRC Press, 2015), pp. 73–97 parameters derived from photoplethysmography and electrocardiog-
[4] Schäfer A., Vagedes J.: ‘How accurate is pulse rate variability as an raphy signals’. IEEE 2015 37th Annual Int. Conf. on Engineering in
estimate of heart rate variability?: a review on studies comparing Medicine and Biology Society (EMBS), 2015, pp. 5952–5955
photoplethysmographic technology with an electrocardiogram’, [14] Pomeranz B., Macaulay R.J., Caudill M.A., ET AL.: ‘Assessment of
Int. J. Cardiol., 2013, 166, (1), pp. 15–29 autonomic function in humans by heart rate spectral analysis’,
[5] Shah S.A., Velardo C., Farmer A., ET AL.: ‘Exacerbations in Am. J. Physiol. – Heart Circulatory Physiol., 1985, 248, (1),
chronic obstructive pulmonary disease: identification and prediction pp. H151–H153
using a digital health system’, J. Med. Internet Res., 2017, 19, (3), [15] Bolanos M., Nazeran H., Haltiwanger E.: ‘Comparison of heart rate
p. e69 variability signal features derived from electrocardiography and
[6] Orphanidou C., Bonnici T., Charlton P., ET AL.: ‘Signal quality indices photoplethysmography in healthy individuals’. IEEE 28th Annual
for the electrocardiogram and photoplethysmogram: derivation and Int. Conf. on Engineering in Medicine and Biology Society
applications to wireless monitoring’, IEEE J. Biomed. Health Inf., (EMBS’06, 2006), 2006, pp. 4289–4294
2015, 19, (3), pp. 832–838
[16] Lin W.-H., Wu D., Li C., ET AL.: ‘Comparison of heart rate variability
[7] Han H., Kim J.: ‘Artifacts in wearable photoplethysmographs during
from PPG with that from ECG’. The Int. Conf. Health Informatics,
daily life motions and their reduction with least mean square based
2014, pp. 213–215
active noise cancellation method’, Comput. Biol. Med., 2012, 42,
[17] Lu G., Yang F., Taylor J.A., ET AL.: ‘A comparison of photoplethys-
(4), pp. 387–393
[8] Han H., Kim M.-J., Kim J.: ‘Development of real-time motion artifact mography and ECG recording to analyse heart rate variability in
reduction algorithm for a wearable photoplethysmography’. IEEE healthy subjects’, J. Med. Eng. Technol., 2009, 33, (8), pp. 634–641
29th Annual Int. Conf. on Engineering in Medicine and Biology [18] Russoniello C.V., Pougtachev V., Zhirnov E., ET AL.: ‘A measurement
Society (EMBS, 2007), 2007, pp. 1538–1541 of electrocardiography and photoplethesmography in obese children’,
[9] Lee C.M., Zhang Y.T.: ‘Reduction of motion artifacts from photo- Appl. Psychophysiol. Biofeedback, 2010, 35, (3), pp. 257–259
plethysmographic recordings using a wavelet denoising approach’. [19] Selvaraj N., Jaryal A., Santhosh J., ET AL.: ‘Assessment of heart rate
IEEE 2003, Medicine and Biology Society (EMBS), Asian-Pacific variability derived from finger-tip photoplethysmography as com-
Conf. on Biomedical Engineering, 2003, pp. 194–195 pared to electrocardiography’, J. Med. Eng. Technol., 2008, 32, (6),
[10] Lee J., Matsumura K., Yamakoshi K.-i., ET AL.: ‘Comparison between pp. 479–484
red, green and blue light reflection photoplethysmography for heart [20] Teng X.F., Zhang Y.T.: ‘Study on the peak interval variability of
rate monitoring during motion’. IEEE 2013 35th Annual Int. Conf. photoplethysmogtaphic signals’. 2003 IEEE Engineering in
on Engineering in Medicine and Biology Society (EMBS), 2013, Medicine and Biology Society (EMBS) Asian-Pacific Conf. on
pp. 1724–1727 Biomedical Engineering, 2003, pp. 140–141
[11] Stein P.K., Kleiger R.E., Rottman J.N.: ‘Differing effects of age on [21] Peng R.-C., Zhou X.-L., Lin W.-H., ET AL.: ‘Extraction of heart rate
heart rate variability in men and women’, Am. J. Cardiol., 1997, variability from smartphone photoplethysmograms’, Comput. Math.
80, (3), pp. 302–305 Meth. Med., 2015, 2015, Article ID 516826, pp. 1–11

64 Healthcare Technology Letters, 2018, Vol. 5, Iss. 2, pp. 59–64

This is an open access article published by the IET under the doi: 10.1049/htl.2017.0039
Creative Commons Attribution -NonCommercial License (http://
creativecommons.org/licenses/by-nc/3.0/)